Scholarlink

Background <p>This study addresses the core challenge in traditional active suspension control systems: the difficulty in synergistically optimizing ride comfort and handling stability.</p> Purpose <p>To tackle this issue, a safety-constrained reinforcement learning control strategy is proposed to achieve multi-objective performance enhancement.</p> Methods <p>By integrating physical safety constraints—including suspension dynamic deflection limits and tire dynamic load ranges—into the reinforcement learning framework, a novel Twin-Delayed Deep Deterministic Policy Gradient algorithm with safety boundary constraints (S-TD3) is designed. The algorithm combines a task-driven multi-objective reward function with a continuous soft-constraint reward mechanism, enabling the derivation of an optimal active suspension control policy through offline training. A comparative experimental framework is established on the MATLAB/Simulink platform, incorporating a Genetic Algorithm-optimized Linear Quadratic Regulator (GA-LQR) and the standard TD3 algorithm for benchmarking.</p> Results <p>Experimental results demonstrate that the S-TD3 algorithm significantly outperforms the GA-LQR algorithm in all evaluated metrics. The proposed method not only enhances vehicle ride comfort but also ensures robust handling stability while reducing suspension dynamic deflection.</p> Conclusion <p>The S-TD3 algorithm effectively addresses the challenge of synergistically optimizing ride comfort and handling stability in active suspension systems, offering a significant improvement over existing methods.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Reinforcement Learning-Based Active Suspension Control Strategy with Safety Constraints

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • Zongwei Zhu,
  • Likang Yang,
  • Sheng Luo,
  • Xinlei Jin,
  • Siliang Fan,
  • Qiyao Zhu

摘要

Background

This study addresses the core challenge in traditional active suspension control systems: the difficulty in synergistically optimizing ride comfort and handling stability.

Purpose

To tackle this issue, a safety-constrained reinforcement learning control strategy is proposed to achieve multi-objective performance enhancement.

Methods

By integrating physical safety constraints—including suspension dynamic deflection limits and tire dynamic load ranges—into the reinforcement learning framework, a novel Twin-Delayed Deep Deterministic Policy Gradient algorithm with safety boundary constraints (S-TD3) is designed. The algorithm combines a task-driven multi-objective reward function with a continuous soft-constraint reward mechanism, enabling the derivation of an optimal active suspension control policy through offline training. A comparative experimental framework is established on the MATLAB/Simulink platform, incorporating a Genetic Algorithm-optimized Linear Quadratic Regulator (GA-LQR) and the standard TD3 algorithm for benchmarking.

Results

Experimental results demonstrate that the S-TD3 algorithm significantly outperforms the GA-LQR algorithm in all evaluated metrics. The proposed method not only enhances vehicle ride comfort but also ensures robust handling stability while reducing suspension dynamic deflection.

Conclusion

The S-TD3 algorithm effectively addresses the challenge of synergistically optimizing ride comfort and handling stability in active suspension systems, offering a significant improvement over existing methods.